KimRass / CLIP Public

Notifications You must be signed in to change notification settings
Fork 0
Star 6

PyTorch implementation of 'CLIP' (Radford et al., 2021) from scratch and training it on Flickr8k + Flickr30k

Name		Name	Last commit message	Last commit date
Latest commit History 139 Commits
configs		configs
data_augmentation_samples		data_augmentation_samples
db		db
papers		papers
.gitignore		.gitignore
README.md		README.md
clip.py		clip.py
data_augmentation.py		data_augmentation.py
db.py		db.py
evaluate.py		evaluate.py
flickr.py		flickr.py
imagenet1k.py		imagenet1k.py
imagenet1k_classes.json		imagenet1k_classes.json
linear_classification.py		linear_classification.py
loss.py		loss.py
model.py		model.py
semantic_search.py		semantic_search.py
set_env.sh		set_env.sh
train.py		train.py
utils.py		utils.py
zero_shot_classification.py		zero_shot_classification.py

Repository files navigation

CLIP (Radford et al., 2021) implementation from scratch in PyTorch

Learning Transferable Visual Models From Natural Language Supervision

Pretrained Model

CLIP trained on Flickr8k + Flickr30k for 200 epochs
- clip_flickr.pth

Zero-shot Classification on ImageNet1k Dataset

# e.g.,
python3 zero_shot_classification.py\
    --data_dir="/Users/jongbeomkim/Documents/datasets/imagenet-mini"\
    --ckpt_path="/Users/jongbeomkim/Documents/clip/checkpoints/clip_flickr.pth"\
    --batch_size=16\
    --n_cpus=4\ # Optional
    --max_len=128\ # Optional
    --k=10 # Optional